[llvm] [AMDGPU] Switch V_CNDMASK operands to shrink it into VOP2 (PR #135162)

[Mirko Brkušanin via llvm-commits]

================
@@ -1010,6 +1162,9 @@ bool SIShrinkInstructions::run(MachineFunction &MF) {
             // provide a hint to the register allocator to use VCC and then we
             // will run this pass again after RA and shrink it if it outputs to
             // VCC.
+            Register NewVCC = trySwapCndOperands(MI);
+            DstReg = NewVCC == AMDGPU::NoRegister ? DstReg : NewVCC;
----------------
mbrkusanin wrote:

Also, why don't you just have trySwapCndOperands return old DstReg instead and have this just be:

DstReg = trySwapCndOperands(MI);

https://github.com/llvm/llvm-project/pull/135162